Purpose-bound delegation

A three-minute film: how the federation, a small language model and PingAuthorize keep a customer's intent bound through a chain of agents.

Explainer film

A customer's intent, kept through a chain of agents

Alice asks one agent to pay her bills. That agent hands parts of the job to others. The film follows the trust model that keeps every hand-off bound to what she asked for: a federation whose superiors vouch for each member's words, a small language model that reads those words in banking's own vocabulary, and PingAuthorize deciding every hop. The scenarios are the live demonstration's; the drawing is an illustration.

0:00

Chapters

    What's real and what's illustrated

    From the live demonstration: every scenario shown, the decisions and the rules that made them, the model's readings and the quotes it gave (ministral-3:3b through Ollama), the purpose map, the entity records and who vouches for them, and the comparison against the delegator alone. You can run each one from the demo and read the recorded runs on Use cases.

    Illustrated: the drawing itself. The federation, the chain and the tokens are shown as a diagram, not as the demonstration's own screens, and the timing is the film's.

    Not covered: the eight further scenarios (borrowing to pay, the bank's own marketing, a helpful extra, an agent for everything, and more), PingFederate as the authorization server, and how quickly revocation takes effect where reliers cache statements.

    Download the film (MP4, 1080p) · How it works has the architecture and the rules in full.

    Script